AI can now tell your genetic risk for health outcomes like heart disease and depression

Genomics is a life science, but machine learning plays an enormous role in its advancement. After all, the human genome has more than 3-billion base pairs of nucleotides. That’s a lot of genetic code to analyze.

Scientists typically use machine learning to synthesize and simplify these massive data sets so researchers can get to what they’re looking for faster.

Polygenic risk scores give an excellent example of how machine learning is helping us understand our health and genome better than ever before, by taking databases with literally trillions of pieces of information and aggregating the most useful data into a single number. That number represents the risk of a person facing a particular health outcome based on their genetics.

These scores are part spreadsheet, part statistics and a little bit of fortune telling. So what place, if any, do they have in our medical care? Let’s explore. 

What is Polygenic Risk?

One concept we reiterate here on Shareable Science quite frequently is that most health outcomes that you and your doctor address don’t boil down to the work of a single gene. If a single gene being flipped on or off does cause a disease, it’s called a Mendelian disorder. However, mostly with human health, we’re talking about what are called complex disorders, because they have many contributing parts.

One complex disorder you hear a lot about is depression, which we talked about in an earlier post: 

Instead of the ‘light switches’ we find with Mendelian disorders, you might be better off thinking about the genetics of complex diseases like depression as a mixing board in a recording studio. It’s not just whether a gene is ‘on’ or ‘off’. It’s also how ‘loud’ or ‘soft’ the gene is. If a gene is heavily over- or underexpressed, it can impact the function of an entire system.

There are hundreds of genes associated with mental health, which you can think of as hundreds of sliders on the mixing board. When you add in environmental factors like upbringing, the inputs get even more complex.

That’s where polygenic risk scores come in, because they can take one look at the whole mixing board, the instruments in the studio and the musicians manning them and predict with some degree of accuracy what kind of sound you’re most likely to get. 

Polygenic risk scores are developed through machine learning. Huge datasets with thousands of genomic variants tested across tens of thousands of individuals are used to train algorithms. The algorithms learn to identify patterns, then compare those patterns to reported health outcomes to create risk profiles. Then the algorithm is validated on a different population. By comparing risk profiles to an individual’s genetic variants, the algorithm can assign that person a polygenic risk score, identifying the likelihood of the health outcome.

Polygenic risk scores have been created for everything from obesity to heart disease, diabetes to depression. They take an immensely complex constellation of data and convert it into a single datapoint that is easy for the average person to understand.

For example, consider a paper published earlier this year describing a polygenic risk score for breast cancer. The score utilizes data from roughly 300 regions of the genome. Women with a polygenic risk score in the top percentile of score distribution have a 33% lifetime risk of developing breast cancer. That’s a valuable piece of information, aggregated from hundreds of individually miniscule data points. It’s likely in the not-too-distant future that a polygenic risk score, rather than a woman’s age, will determine when she should begin mammography-based screening.

What Polygenic Risk Scores Offer

Healthcare professionals see immense potential in these scores because of their ability to help with personalizing prevention, early detection and treatment. 

Sometimes small lifestyle changes can prevent large health complications down the road, leading to better outcomes and lower costs for everyone. If you have a high polygenic risk score for heart disease, you might change your diet before you start to see symptoms, which could address the problem before it becomes a real issue. Likewise, someone with a high polygenic risk score for depression might be more sensitive to emotional struggles and more attentive to their mental health.

These scores are especially relevant for people with limited family health history, like adoptees. Family history and background can be incredibly important for physicians as they craft your care plan. If cancer runs in your family, doctors may screen more aggressively and at a younger age. If both your parents deal with bipolar disorder, a doctor may ask more specific questions about your mental health.

Without that family history, people can’t tailor their lifestyles to their individual risks, making it far less likely that they prevent or minimize an otherwise-avoidable health outcome. Polygenic risk scores help to fill in some of the blanks that you might find due to a lack of family history, and the scores can also provide lots of additional context for people who do have that history available to them.

Putting the Risk in Polygenic Risk

One of the most significant pitfalls associated with polygenic risk scores is their current bias towards individuals of European descent. This stems from a lack of diversity in the populations initially recruited for these large-scale genomic studies. As a result, polygenic risk scores can mis-estimate risk when applied to non-European populations. An obvious solution is to increase the participation of diverse populations in genomic studies, but this will take time. 

There are also concerns about what introducing polygenic risk scores to patients might mean. For example, what evidence do we have that knowing this information improves patient care? How is a person’s life impacted if they live in fear of an outcome that never comes to pass? Additional studies are needed to weigh the potential benefits and harms of using these scores for clinical decision-making. Medical practitioners ultimately work to serve the patient, so understanding the impact of the score on a patient’s life is an important part of the equation.

It’s also possible that people might use the data to justify a bad health decision or ignore early warning signs of a disease because they believed that a lower than average polygenic risk score meant they were completely protected. It’s important to recognize that even the best algorithms don’t account for all risk factors. A low genetic risk of heart disease doesn’t give a green light for eating high-fat foods around the clock. 

A Genomically Literate Society

The opportunities and concerns generated by polygenic risk scores all point to one thing—the need for a genomically literate society. More than ever, we are able to use genetic data to inform health decisions, but it only works if people correctly understand the information they’re receiving.

People still have an enormous amount of control over their health outcomes, and understanding genetic predisposition makes that more true, not less. 

To schedule a media interview with Dr. Neil Lamb or to invite him to speak at an event or conference, please contact Margetta Thomas by email at mthomas@hudsonalpha.org or by phone: Office (256) 327-0425 | Cell (256) 937-8210